Botsitting: Conceptualising Value-Enabling Hidden Human Labour in AI-Enabled Work

Citation: Dwivedi, Y.K., Crick, T., Michael, K., Samothrakisn, S., Sarker, I., Naeema, R.M., Raman, R., Malik, T., Dutota, V., Jeon, Il, Helal, M.Y.I., 2026, “Botsitting: Conceptualising Value-Enabling Hidden Human Labour in AI-Enabled Work”, Journal of Computer Information Systems, https://doi.org/10.1080/08874417.2026.2749411

Photo by Mikhail Nilov

Abstract

Artificial intelligence (AI) is transforming organisational work by creating unprecedented opportunities to augment human capability and generate organisational value. Yet AI-generated potential does not automatically become realised value, highlighting a fundamental challenge concerning how organisations transform technological capability into meaningful outcomes. Responding to this challenge, this paper advances, to the best of our knowledge, one of the first scholarly conceptualisations of Botsitting as an emerging form of value-enabling hidden human labour. In doing so, it develops an integrated conceptual model explaining Botsitting's contribution to AI value realisation, advances theoretically reasoned research propositions, and proposes a structured future research agenda. The paper contributes to AI-enabled work, IT business value, and human-AI collaboration by explaining one previously under-conceptualised pathway through which AI-generated potential becomes realised organisational value.

Keywords: Botsitting, AI value realisation, artificial intelligence, hidden human labour, AI-enabled work, human-AI collaboration, productivity paradox.

Introduction

Artificial intelligence (AI) has rapidly become one of the defining technologies shaping contemporary organisations. Advances in generative AI and, increasingly, agentic AI are transforming how individuals, teams, and organisations generate knowledge, make decisions, solve problems, and create value. Across industries, organisations continue to invest heavily in AI-enabled work with expectations of improving productivity, enhancing decision quality, increasing operational efficiency, stimulating innovation, and strengthening competitive advantage1,2,3,4,5,6. At the same time, AI is increasingly viewed not simply as a technological innovation but as a catalyst for redesigning organisational work through new forms of human-AI collaboration and hybrid intelligence7,8,9. Consequently, scholarly and managerial attention has progressively shifted from whether organisations should adopt AI, to how AI capabilities can be effectively integrated into organisational work to realise sustainable value through the augmentation of human capability.

Despite these expectations, the relationship between AI capability and organisational value remains considerably more complex than prevailing technological narratives often imply. Although contemporary AI systems are capable of generating sophisticated analyses, recommendations, content, decisions, and increasingly autonomous actions, these capabilities do not automatically translate into valuable organisational outcomes. Organisations frequently discover that AI-generated outputs require human interpretation, contextualisation, verification, refinement, and professional judgement before they can be confidently incorporated into organisational work10,7. A July 2025 study conducted by MIT revealed that despite a $30-$40 billion investment into generative AI, just 5% of these generative AI projects are generating returns11. Rather than eliminating human work, increasing AI capability frequently redistributes it by creating new forms of complementary human contribution that remain essential in the AI-augmented landscape for achieving reliable organisational outcomes. Though one can think of this as a paradox, there are certain links to labour theory of value12 that are hard to ignore. Namely, the creation of value seems to require, or to be intrinsically linked to, human labour, not necessarily in a causal setup (i.e. labour causing value) but rather as an inescapable component of any AI-enabled value proposition.

This challenge reflects a broader and enduring concern within Information Systems (IS) research. Decades of scholarship on the productivity paradox and IT business value have consistently demonstrated that technological capability alone does not create organisational value13,14,15. Instead, technology creates opportunities for value, whereas realised value depends upon complementary organisational resources, governance arrangements, managerial capabilities, work practices, and human expertise. Recent research on AI similarly argues that organisational value increasingly emerges through augmentation rather than substitution, where AI capability and human judgement complement one another in organisational decision making and knowledge work16,17,8,18,2. The fundamental question therefore is no longer whether AI is capable of generating increasingly sophisticated outputs, but how AI-generated capability becomes transformed into reliable, contextually appropriate, and organisationally valuable outcomes19.

Although this broader principle has long been recognised, AI-enabled work introduces a distinctive manifestation of the value realisation challenge. Unlike previous generations of digital technologies, contemporary AI systems actively participate in organisational work by generating recommendations, content, analyses, decisions, and increasingly autonomous actions. Yet these outputs frequently remain provisional rather than immediately actionable. Their practical value often depends upon additional human effort to evaluate, interpret, validate, refine, contextualise, and integrate them into specific organisational settings. Furthermore, workslop can be created, referring to the AI-enabled polishing of documents that often lack substance and accuracy, frequently creating additional work for those required to read and interpret them. This results in additional human correction and rework20. Therefore, advances in AI capability simultaneously create unprecedented opportunities for organisational value creation while increasing the importance of complementary forms of human labour required to realise that value.

We describe this emerging challenge as the “AI Value Realisation Paradox”. The paradox does not suggest that AI fails to create value, nor does it question the transformative potential of AI-enabled work. Rather, it holds that as AI capability increases, organisations may generate ever more technological potential while continuing to depend upon complementary human work to convert that potential into meaningful outcomes. Understanding that work is therefore central to explaining how organisations realise value from AI-enabled work.

Interestingly, while academic understanding of this complementary human work remains comparatively limited, practitioner and professional communities have begun to recognise it explicitly. The Glean Work AI Index names the phenomenon Botsitting, reporting that many knowledge workers spend significant time overseeing and correcting AI-generated work rather than simply consuming AI outputs21. The same source identifies a corresponding failure mode, “botshitting”, in which the effort of oversight comes to feel unrewarded and AI-generated outputs are released without proper review. Comparable observations have appeared across enterprise technology and business publications22,23,24,25. Collectively, these observations suggest that complementary human labour has become a visible feature of organisational AI deployment rather than an exceptional organisational practice.

The emergence of the term Botsitting, however, should not be confused with the emergence of the underlying organisational phenomenon. Individuals have long reviewed, interpreted, validated, corrected, and contextualised outputs generated by expert systems, decision support systems, machine learning applications, analytics platforms, and other intelligent technologies. What is new is not the existence of these complementary forms of human work, but their increasing visibility within AI-enabled work and the emergence of a practitioner vocabulary through which they can be collectively recognised and discussed. Distinguishing between the organisational phenomenon and the language used to describe it is therefore essential for rigorous concept development.

This distinction reflects a broader characteristic of knowledge development. Organisational practice frequently recognises emerging phenomena before scholarship develops the conceptual foundations required to explain them. Practitioner communities identify recurring organisational patterns, assign labels, and communicate practical experience, whereas scholarly conceptualisation seeks to explain the underlying nature of those phenomena, establish their conceptual boundaries, distinguish them from related concepts, and position them within existing theoretical knowledge26,27,28,29.

Responding to this opportunity, this paper advances, to the best of our knowledge, one of the first scholarly conceptualisations of Botsitting. Rather than conceptualising Botsitting simply as the supervision of AI systems, we argue that it represents an emerging form of value-enabling hidden human labour that contributes to transforming AI-generated potential into realised organisational value. Botsitting is not proposed as the sole explanation for AI value realisation, nor does it replace existing theories of IT business value, socio-technical systems, organisational capabilities, or human-AI collaboration. Instead, it provides a complementary conceptual lens through which one previously under-theorised pathway linking AI-generated potential and realised organisational value can be more clearly understood.

Against this background, this paper addresses the following research question: How can Botsitting be conceptualised as an emerging form of value-enabling hidden human labour that contributes to AI value realisation in AI-enabled work?

To address this question, the paper pursues three objectives, which correspond to its three contributions. First, it examines the emergence of Botsitting as an organisational phenomenon by synthesising early practitioner and professional evidence, distinguishing between the organisational phenomenon, the practitioner terminology, and the scholarly concept. Second, it advances one of the first scholarly conceptualisations of Botsitting by establishing its ontological identity, conceptual definition, boundaries, defining attributes, and functional role, thereby establishing its conceptual identity within the emerging literature on AI-enabled work. Third, it develops an integrated conceptual model explaining how Botsitting contributes to narrowing the AI Value Realisation Gap, together with theoretically reasoned propositions and a structured future research agenda intended to support cumulative theory development and subsequent empirical investigation.

The remainder of the paper is organised as follows. Section 2 outlines the concept development approach adopted to identify and synthesise early practitioner and professional evidence. Section 3 examines the emergence of Botsitting as an organisational phenomenon and Section 4 develops its scholarly conceptualisation. Building upon this conceptual foundation, Section 5 presents an integrated conceptual model of Botsitting and AI value realisation. Sections 6 and 7 advance emerging research propositions and a structured future research agenda, respectively. Section 8 discusses the theoretical and practical implications of the conceptualisation before the paper concludes by reflecting on the broader significance of understanding the complementary human labour through which AI-generated potential becomes realised organisational value.

Research Method

This paper adopts a concept development approach to advance a scholarly conceptualisation of Botsitting as an emerging organisational phenomenon. Rather than conducting a conventional systematic literature review, the objective is to develop a theoretically grounded conceptualisation capable of explaining the nature, boundaries, and contribution of Botsitting to AI-enabled work. Such an approach is appropriate because Botsitting has only recently emerged within practitioner and professional discourse, while the academic literature has yet to establish a coherent conceptual foundation. Concept development is particularly valuable when organisational practice evolves more rapidly than scholarly theorisation, requiring researchers to transform emerging phenomena into theoretically meaningful concepts30,26,27,29. This responsiveness to practice reflects a broader concern with narrowing the gap between research and organisational reality31.

The methodological approach adopted in this study reflects the complementary roles of practitioner and professional evidence and established academic theory. Following Webster and Watson’s32 view that practical experience and emerging phenomena can provide important sources for theory development, this paper draws on the emerging practice of Botsitting to develop a conceptual understanding of the human work required to translate AI capabilities into organisational value. Practitioner and professional sources provide early evidence of emerging organisational phenomena by documenting recurring practices, shared terminology, and evolving organisational experience. Such evidence helps identify and characterise the phenomenon, but it does not, by itself, provide a scholarly explanation. The academic literature provides the theoretical foundations through which the phenomenon can be interpreted, conceptualised, and positioned within existing knowledge. Accordingly, practitioner and professional evidence was used to identify and characterise Botsitting, whereas scholarship on concept development, AI-enabled work, IT business value, socio-technical systems, human–AI collaboration, and hidden human labour informed its conceptualisation.

The Emergence of Botsitting: From Organisational Phenomenon to Scholarly Concept

The emergence of Botsitting should be understood as the increasing visibility of an organisational phenomenon rather than the appearance of an entirely new form of work. As AI has become progressively embedded within organisational processes, knowledge work, and professional practice, the reality is that human involvement has not diminished. Rather, the nature of human work has evolved, with AI augmenting work rather than simply replacing it33. Revisiting this fundamental understanding is timely given that generative AI and agentic AI are being perceived by industry as a labour substitute, with task-level and role-level substitution featuring more strongly than complete workforce replacement34,35. However, across a wide range of AI-transformed organisational contexts, individuals are increasingly devoting time and expertise to supervising, interpreting, validating, refining, correcting, and contextualising AI-generated outputs before those outputs can be relied upon in practice. Although these activities are frequently embedded within routine organisational work and therefore receive limited explicit recognition, they have become an increasingly important component of successful AI-enabled work10,7,8. Consistent with this emerging pattern, a 2026 report from Glean suggests that employees spend approximately 6.4 hours per week on Botsitting-related activities21. Because this evidence originates from a commercial provider of enterprise AI tooling, it is best read as indicative rather than definitive. Independent experimental evidence nonetheless points in a compatible direction. Controlled studies report substantial generative AI productivity gains on well-specified tasks36, yet field experimental work shows those gains reversing on tasks falling outside the model's competence, where participants who accepted AI output uncritically performed worse than those working unaided37. A meta-analysis of 106 experiments further finds that human-AI combinations on average underperform the better of human or AI alone, with the quality of human verification a central moderator38. Together these findings support the premise developed here while indicating that the magnitude of Botsitting effort remains an open empirical question.

This reflects a broader pattern throughout the history of technological change: innovations rarely eliminate human work entirely, but redistribute it. The adoption of CAD systems, for example, generated new supporting roles and forms of expertise alongside the automation of design activities, created as needed to support the newly automated toolchain39. Contemporary AI is a particularly significant manifestation of this pattern. As AI systems generate increasingly sophisticated analyses, recommendations, content, and autonomous actions, organisations become correspondingly more dependent upon the human judgement that ensures those outputs are reliable and contextually appropriate. Increasing AI adoption therefore does not simply automate existing work; it makes previously overlooked forms of human labour more visible and integral to successful outcomes.

This redistribution of work is already evident across professional occupations where AI is reshaping how tasks are performed and coordinated. In marketing, this transition is particularly pronounced. Although a large proportion of marketing-related tasks are automatable, reshaping the structure of early-career and analytical work, firms are not eliminating marketing as a function40. Instead, they are reorganising how marketing activities are produced. AI increasingly absorbs routine analytical and content-generation tasks, reducing demand for some junior execution roles while increasing the importance of workers who can effectively collaborate with AI systems. These emerging roles require employees to interpret model outputs, translate AI-generated insights into strategy, manage AI-supported campaigns, and oversee automated workflows. Consistent with this shift, demand for AI-related skills in sales and marketing roles has increased substantially41. The implication is that marketing work is not disappearing but becoming more technical, strategic, and AI-augmented, with greater emphasis on hybrid capabilities that combine domain expertise with AI literacy.

In banking, a similar restructuring is evident, with AI transforming work around new forms of human-AI coordination. Senior leadership within major financial institutions has emphasised that AI will affect virtually all categories of work in the sector42. Rather than eliminating banking work, AI is expected to reduce demand for some traditional roles while increasing demand for positions involving AI development, oversight, and integration. Banking work is being reorganised around AI-enabled workflows in which routine analytical and coordination activities are automated, while human roles concentrate in oversight, client-facing judgement, regulatory compliance, and strategic decision making. In practice, this effort involves evaluating AI-generated outputs, resolving ambiguity, exercising contextual judgement, identifying errors, and integrating recommendations into organisational processes. These are not exceptional interventions undertaken only when AI performs poorly; they are routine practices through which AI-generated potential becomes reliable and viable in specific organisational contexts.

In physical security, a comparable reconfiguration is visible, with AI reshaping the role of closed-circuit television (CCTV) operators from passive screen watchers into active responders43. Traditionally, monitoring a dozen live feeds in a security operations centre (SOC) for extended periods produced a form of vigilance decrement, with attentiveness degrading through hours of low-stimulus observation. AI-based video analytics now absorb much of this continuous visual scanning, filtering out routine motion and alerting operators only to context-rich events such as loitering, falls, or unauthorised access44,45. Operators are thereby repositioned as decision-makers who intervene proactively rather than watch continuously46, supported by natural language interfaces that allow footage to be queried conversationally rather than manually reviewed47,48, while AI's capacity to sustain analysis across many feeds simultaneously has increased the camera-to-operator ratio without a corresponding loss of monitoring effectiveness43.

Yet this transition carries its own paradox49. Where the passivity of screen watching produced fatigue through under-stimulation, remaining on standby to be summoned by an AI system at any moment may generate a different kind of fatigue rooted in intermittent, high-alert readiness. Redistributing cognitive load onto AI does not eliminate operator fatigue so much as transform its underlying mechanism, from passive vigilance decrement to reactive alert strain. The risk that oversight becomes nominal is not hypothetical: investigations into UK CCTV provision found that although cameras operated constantly and everywhere, no one was reliably watching in the security operations centre50. This raises an open question of when automated monitoring and human-in-the-loop alerting become unmanageable, with operators facing a volume of AI-generated alerts no human could realistically oversee, a limit examined in Section 5.3.

Recent research similarly argues that AI reshapes rather than replaces organisational work, with human expertise complementing AI capability through interpretation, contextualisation, ethical judgement, and organisational decision making16,7,3. Scholarship on hidden and invisible forms of labour offers a further perspective. Research on invisible work, ghost work, and hidden human infrastructure demonstrates that apparently autonomous digital technologies frequently depend upon human labour that remains organisationally under-recognised despite its importance51,52,53,54,55. That literature has primarily examined the labour involved in developing, training, and moderating AI systems; the present paper turns instead to the complementary human labour through which AI-generated potential becomes realised organisational value during organisational use.

A classic analogy for Botsitting is that of babysitting, or more broadly, childcare. Table 1 draws on this analogy to illustrate the significance society has traditionally placed on raising children toward human flourishing, laying out fourteen comparative dimensions that capture the depth of care involved when an employee engages in what is often characterised as routine “AI oversight labour” but is, in effect, something considerably more demanding. Childcare is rarely described as “just” childcare, and is seldom classified as “work” in the conventional sense, yet much about the act resonates with Botsitting. While admittedly contestable, employees today are not simply “watching” bots or agents; they are, in a meaningful sense, helping to raise them. Where humans once sowed seeds in the earth, they now cultivate bots, nurturing them in the form of algorithms and autonomous agents.

Table 1. Comparing the Human Act of Childcare vis-à-vis Botsitting

Dimension

Childcare

Botsitting

Nature of the activity

Continuous watching, feeding, soothing, responding to needs before they escalate

Continuous monitoring, prompting, correcting, and validating AI outputs before errors escalate

Visibility of effort

Largely invisible, happens in the home, rarely counted as "work"

Largely invisible, happens alongside "real" tasks at work, rarely logged as distinct labour

Economic recognition

Historically unpaid or underpaid; excluded from GDP calculations

Rarely budgeted for; absorbed into existing job roles without additional compensation or training

Perceived skill level

Often framed as "natural" or instinctive rather than skilled labour

Often framed as "just double-checking" rather than a distinct competency such as validating, refining, contextualising, or correcting

Temporal pattern

Unpredictable, intermittent demands (a baby cries without warning)

Unpredictable, intermittent demands (a model output needs correction without warning)

Cognitive load

Sustained vigilance punctuated by sudden high-alert moments (e.g., illness)

Sustained vigilance punctuated by sudden high-alert moments (reviewing flagged AI outputs and exceptions)

Fatigue type

Vigilance fatigue from being “on call” even during quiet periods (e.g., night-time)

Vigilance fatigue from being “on call” even when the AI is performing well (something can go wrong when the worker takes a quick break)

Institutional framing

Categorised as a private, domestic responsibility rather than professional work that could be certified

Categorised as an informal add-on to a job description rather than a distinct role (a default expectation that one can use AI and work faster)

Who bears the cost

Disproportionately falls on women, unpaid family members, and carers

Disproportionately falls on junior staff or those without formal authority to push back. Greater expectations are placed on operational staff

Outcome when done well

Child thrives but the caregiver's labour is credited to the child's “natural development”

AI output is reliable but the labour is credited to the AI system's “intelligence” and not the worker's “intelligence”

Outcome when done poorly

Harm is highly visible and attributed to caregiver failure

Harm (errors, bias, hallucinations) is visible but often attributed to the AI, not the absent oversight

Measurement in official statistics

Excluded from labour statistics and productivity metrics; the caregiver receives no monetary compensation for this labour save for possible social security rebates dependent on household income

Excluded from AI productivity and ROI metrics

Language used to describe it

“Just watching the baby”, “naturally caring”, “every parent has the responsibility to do it, because they are accountable for their child”

“Just double-checking”, “keeping an eye on it”, “make sure nothing goes wrong”, “the output has to stand up to the stakeholder’s expectations”, “the sources need to be correct”

Structural consequence

Undervaluing childcare masks its necessity for a functioning society; poor care usually means poor societal mental health statistics and lower-than-expected workforce productivity

Undervaluing Botsitting masks its necessity for AI systems to generate real organisational value; creating poor conditions for improving LLM quality and risking a downward spiral in what is known as the “curse of dimensionality”19

The marketing, banking, and CCTV examples discussed above share a common but underexamined feature: in each case, the realisation of AI's technological potential depends on a form of human labour that is continuous, reactive, and largely uncredited. Table 1 sets out why childcare offers an instructive parallel. The exclusion of care work from formal labour statistics is not incidental; it reflects a broader societal tendency to under-value work that is intermittent, reactive, and difficult to observe from the outside. A caregiver may appear idle during quiet moments yet remains in a state of continuous readiness, and the same asymmetry of credit and blame recurs in both settings: when things go well the outcome is attributed elsewhere, and when they go wrong the oversight failure becomes immediately and disproportionately visible.

Another relevant analogy for Botsitting can be found in the long-standing literature on human vigilance in automated and semi-automated systems, the kind of labour performed by air traffic controllers, nuclear and chemical plant control room operators and pilots monitoring autopilot systems. Mackworth's (1948) wartime studies of radar operators, which first identified the vigilance decrement, showed that sustained attention to a low-stimulus monitoring task degrades measurably over time, even though the task appears passive or effortless from the outside. Botsitting exhibits the same attributes: long stretches of low-intensity output-checking punctuated by sudden, high-stakes moments of intervention when a model's output drifts, hallucinates, or fails silently. The same dynamic recurs in AI-enabled work, and its implications for the limits of Botsitting are examined in Section 5.3.

Read together, these parallels suggest that the AI value realisation gap identified throughout this paper is not simply a technical or organisational oversight, but an instance of a broader societal pattern in which essential, sustained forms of supervisory labour, whether directed at children or at machines, are systematically rendered invisible, uncompensated, and unaccounted for in official measures of value and productivity.

Toward a Shared Vocabulary

As organisations accumulated experience with generative AI and, more recently, agentic AI, these recurring patterns of complementary human work became progressively easier to recognise5,6. Across diverse industries and professional settings, individuals encountered a common organisational reality: AI-generated potential rarely became realised value without meaningful human involvement.

As the complementary human work associated with AI-enabled systems became increasingly visible, practitioner and professional communities gradually developed a shared vocabulary through which the phenomenon could be recognised and discussed. Among the terms that emerged, Botsitting gained particular prominence as a concise term for the human work involved in supervising, monitoring, verifying, refining, correcting, and managing AI-generated outputs and AI-enabled processes. Recent industry research, enterprise technology publications, business media, and practitioner commentary consistently portray Botsitting as an increasingly important, yet frequently overlooked, aspect of organisational AI adoption21,22,23,24,25. Practitioner synthesis of these findings further illustrates the empirical scale and organisational consequences of Botsitting. Drawing on the Work AI Index 2026, SoftwareSeni characterises Botsitting as a widespread form of hidden corrective labour accompanying generative AI adoption and contrasts it with botshitting, where insufficient review leads to downstream rework and risk57. Although the sources discussed differ in emphasis and organisational context, they collectively recognise a common organisational reality. AI-generated potential frequently requires complementary human effort before it becomes organisationally valuable.

The emergence of the term Botsitting marks a stage in the recognition of the phenomenon rather than its origin. This growing discursive visibility should nonetheless be distinguished from organisational recognition in a formal sense: Botsitting may be increasingly discussed within professional and media discourses while remaining largely unbudgeted, unmeasured, and unattributed within organisational value accounts.

Existing practitioner and professional discussions describe Botsitting through observable activities, but outcome checking is particularly important given AI's potential inaccuracies and their consequences. This can include finding plausible but incorrect statements, detecting insensitive answers, spotting misbookings, privacy and security issues, noncompliance, policy violations, hallucinated references, or incorrect actions. There is a need for tools and support for checking outcomes, as checking everything may simply not be feasible given the widespread use of AI and the fact that error detection tends to degrade as cognitive load rises and concentration falls. Incorrect actions may require corrective follow-up such as rerouting tasks, correcting calculations, redoing transactions, changing numbers, or rebuilding tables. AI itself can be used for this and thereby influences how Botsitting is arranged, so both AI and human capabilities must be addressed to enable value creation. Corrective actions may in turn help humans learn to perform Botsitting more effectively. Treating Botsitting as merely a form of quality control is therefore too narrow: the focus should be on creating organisational value, which requires continuous learning across the whole process of managing and improving AI.

The emergence of a shared vocabulary is more than a linguistic development; it often represents the first stage in the institutionalisation of an organisational phenomenon. Once organisations possess a common language to describe recurring practices, they are better positioned to develop governance mechanisms, performance measures, capability frameworks, and managerial interventions. Establishing Botsitting as a recognised concept therefore creates a foundation for both scholarly inquiry and organisational practice.

Recognising this distinction provides the foundation for the remainder of the manuscript. Having established Botsitting as an emerging organisational phenomenon and demonstrated how practitioner communities came to recognise and describe it, the next task is to determine its ontological identity and develop a rigorous scholarly conceptualisation. The following section therefore moves from organisational observation to conceptual explanation by defining what Botsitting fundamentally is, establishing its conceptual boundaries, identifying its defining attributes, and explaining its contribution to AI value realisation.

Scholarly Conceptualisation of Botsitting

Ontological Identity of Botsitting

Having established Botsitting as an emerging organisational phenomenon, the next task is to determine its conceptual identity. Naming a phenomenon does not, by itself, explain what it fundamentally represents. Scholarly conceptualisation requires moving beyond descriptive accounts of organisational experience towards concepts that are theoretically meaningful, conceptually coherent, and capable of supporting cumulative theory development30,26,27,58,29. This section therefore establishes the ontological identity of Botsitting and positions it within the broader literature on AI-enabled work and organisational value realisation.

Conceptualisation begins by explaining the essential nature of a phenomenon rather than cataloguing the activities through which it becomes observable. Observable activities inevitably evolve as technologies, organisational practices, and institutional contexts change. Concepts, however, should possess sufficient stability to support cumulative theoretical development across changing empirical settings26,28. Consequently, Botsitting should not be conceptualised solely through contemporary activities such as monitoring AI systems, verifying outputs, refining prompts, correcting errors, or supervising autonomous agents. These activities are important because they make the phenomenon visible in organisational practice, but they represent manifestations of a deeper underlying organisational phenomenon rather than its conceptual essence.

Building upon the methodological approach adopted in this paper and the dialogue between practitioner and professional evidence and established academic scholarship, Botsitting is conceptualised as an emerging form of value-enabling human labour that frequently remains hidden or under-recognised within organisations.

At this juncture, it is important to clarify the difference between human-in-the-loop (HITL) and Botsitting efforts. The distinction between HITL and Botsitting is best understood as one between specified and unspecified oversight labour. HITL refers to the human involvement deliberately designed into an AI system's functional blueprint. It refers to the verification, validation, and intervention activities that are defined, allocated to identifiable actors, and formally approved before deployment59. Botsitting, by contrast, is the unplanned labour that arises when this specification is absent or incomplete. This effort is improvised by workers since "nobody at deployment defined what verification was required, who owned it, or what good output looks like"60. Thus, whereas HITL is sanctioned, budgeted, and accountable oversight, Botsitting is its unrecognised and unbudgeted remainder. It is the labour that falls outside the approved design and is therefore borne invisibly by the individual workers.

Conceptualising Botsitting as human labour reflects the observation that AI-enabled work frequently depends upon complementary human effort before AI-generated potential becomes organisationally valuable61. Contemporary research increasingly argues that AI capability creates opportunities for augmentation rather than complete substitution, with organisational outcomes depending upon complementary interactions between AI systems and human expertise10,7,8. Botsitting therefore represents a distinctive form of human labour situated within the broader socio-technical processes through which AI-generated potential is transformed into realised organisational value9.

Describing Botsitting as hidden human labour further positions the concept within a broader and rapidly developing body of scholarship examining forms of work that remain organisationally essential despite often receiving limited recognition. Research on invisible work, ghost work, and hidden human infrastructure demonstrates that apparently autonomous digital technologies frequently depend upon complementary forms of human labour that remain organisationally under-recognised51,62,52,55,53,54. Although this literature has primarily examined the human labour involved in developing, training, moderating, and maintaining AI systems, it establishes a broader theoretical principle that is directly relevant to AI-enabled work. Technological capability alone rarely produces valuable outcomes without complementary human effort49.

The conceptualisation advanced in this paper extends this emerging literature in an important way. Existing research has primarily focused on the hidden labour supporting the production and operation of AI systems63. Botsitting, by contrast, focuses on the complementary human labour required during organisational use to transform AI-generated potential into realised organisational value. In this sense, Botsitting should not be understood as another form of ghost work or invisible labour. Rather, it represents a distinct form of hidden human labour whose defining characteristic lies in its contribution to AI value realisation.

Finally, Botsitting is conceptualised as value-enabling because its contribution is fundamentally relational rather than independent. Botsitting does not create organisational value separately from AI capability, nor does it replace existing explanations of IT business value or socio-technical value creation. Instead, it comprises the complementary human labour through which AI-generated potential becomes reliable, contextually appropriate, and organisationally valuable. Conceptualising Botsitting in this way aligns the phenomenon with the broader literature on IT business value while simultaneously explaining why increasing AI capability alone does not guarantee increasing organisational value14,15.

Defining Botsitting: Conceptual Boundaries, Manifestations, and Defining Attributes

Having established the ontological identity of Botsitting, the next stage is to define the concept more precisely. Conceptual definitions provide a shared understanding of what a concept represents while distinguishing it from related concepts. A robust definition should therefore capture the essential nature of the phenomenon without becoming constrained by the particular technologies, activities, or organisational contexts through which it is observed30,26,28. This is especially important in AI-enabled work, where rapid technological development continually reshapes organisational practices while the underlying phenomenon may remain comparatively stable.

Building upon the ontological position established in the preceding section, Botsitting is defined as follows: Botsitting refers to an emerging form of value-enabling, initially unplanned and frequently hidden human labour that contributes to transforming AI-generated potential into realised organisational value through the supervision, monitoring, verification, refinement, interpretation, correction, and management of AI-generated outputs and AI-enabled processes.

The definition deliberately separates the conceptual identity of Botsitting from the activities through which it becomes observable. The first clause establishes the underlying nature of the concept; the second identifies the principal manifestations through which it is currently expressed in organisational practice. Those manifestations are expected to evolve alongside advances in AI capability and organisational AI maturity, but they do not define the concept itself.

The definition also recognises that Botsitting operates across multiple levels of organisational analysis. At the individual level, Botsitting may involve reviewing AI-generated outputs, validating recommendations, refining prompts, or exercising contextual judgement before AI-enabled work is implemented. At the team level, it may encompass collaborative review, peer evaluation, shared decision making, and collective quality assurance. At the organisational level, Botsitting may include governance arrangements, oversight mechanisms, organisational policies, and managerial practices designed to ensure that AI-enabled work remains reliable, responsible, and aligned with organisational objectives within human–AI coordination processes60. At the macro level, Botsitting extends beyond the boundaries of any single organisation to encompass the regulatory frameworks, industry standards, professional accreditation regimes, and institutional norms through which sectors and societies collectively monitor, correct, and legitimise AI-enabled work — reflecting a broader infrastructure of human oversight that underpins trust in AI systems at a systemic level. Although these manifestations differ across analytical levels, they collectively represent the same underlying form of value-enabling hidden human labour.

As organisational AI maturity increases, Botsitting may progressively evolve from an informal and largely invisible activity into an explicitly recognised organisational capability. Similar to quality assurance, cybersecurity, and knowledge management, activities that initially emerge as informal responses to technological change may later become formalised through governance structures, specialised roles, organisational routines, and professional development. Recognising this evolutionary trajectory further reinforces Botsitting's conceptual distinctiveness.

Establishing a conceptual definition also requires clarifying what the concept is not. Botsitting should not be equated with every form of human interaction with AI systems. Routine use of AI applications, casual prompting, or passive acceptance of AI-generated outputs does not necessarily constitute Botsitting. Likewise, Botsitting should not be treated as synonymous with AI governance, digital labour, human-AI collaboration, or organisational management, although it may contribute to each of these broader domains. Botsitting emerges specifically where complementary human labour, without prior explicit planning, contributes to transforming AI-generated potential into reliable, contextually appropriate, and organisationally valuable outcomes through activities that supervise, evaluate, interpret, validate, refine, or otherwise improve AI-enabled work.

A comparable pattern is evident in software engineering, where generating code through natural language prompting (“vibe coding”) requires systematic human review, testing, and correction before deployment64. As AI technologies evolve, some manifestations are likely to diminish while others emerge, and defining Botsitting solely through its current manifestations would tie the concept to a particular stage in the evolution of AI-enabled work.

Table 2. Defining Attributes of Botsitting

Attribute

Description

Human-performed

Requires active human judgement, intervention, and responsibility.

AI-dependent

Emerges within AI-enabled work and would not exist independently of AI systems.

Value-enabling

Contributes to transforming AI-generated potential into realised organisational value.

Context-dependent

Draws upon organisational, professional, and domain-specific knowledge.

Iterative

Frequently involves repeated cycles of evaluation, refinement, and validation.

Frequently hidden

Often remains under-recognised despite contributing substantially to AI value realisation.

Multi-level

Operates across individuals, teams, organisations, and society (micro, meso, macro).

Unspecified

Occurs over and above the human effort already specified at the time of system design to process and render an AI-generated output usable.

Instead, Botsitting is characterised by a set of defining attributes (see Table 2) that remain comparatively stable despite continuing technological and organisational change. These defining attributes distinguish Botsitting from related forms of human work while providing the conceptual stability necessary for cumulative theory development26,29.

Collectively, these defining attributes provide conceptual stability independent of particular technologies, applications, or organisational settings. They distinguish Botsitting according to its enduring characteristics rather than its temporary manifestations, thereby strengthening conceptual clarity and providing a robust foundation for cumulative theoretical development and subsequent empirical investigation.

Functional Role of Botsitting in AI Value Realisation

The conceptualisation so far establishes what Botsitting is. A scholarly concept becomes theoretically meaningful, however, not only by explaining what it is but also why it matters. This section therefore positions Botsitting within the broader process through which AI-generated capability becomes realised organisational value.

The central argument advanced in this paper is that AI capability creates AI-generated potential but does not, by itself, guarantee organisational value. Consistent with the broader literature on IT business value, socio-technical systems, organisational capabilities, and human-AI collaboration, realised value emerges through the interaction of technological capability and complementary human and organisational contributions rather than through technology alone14,15,61,7,8. Within this broader process, Botsitting is conceptualised as one previously under-conceptualised form of value-enabling hidden human labour that contributes to transforming AI-generated potential into reliable, contextually appropriate, and organisationally valuable outcomes.

Botsitting performs this value-enabling role by supplying the complementary human judgement, contextual understanding, verification, interpretation, and refinement required to bridge AI-generated potential and realised value. Rather than positioning AI capability and human expertise as competing alternatives, the conceptualisation recognises that organisational value frequently depends upon their interaction.

Positioning Botsitting in this way also extends the emerging literature on hidden human labour, which has primarily examined the work of producing, training, maintaining, and moderating AI systems52,63,53,54. The present conceptualisation shifts attention towards the hidden labour required after AI systems are deployed within organisational settings: not producing AI capability, but enabling AI-generated potential to become realised organisational value.

Importantly, the conceptualisation does not imply that Botsitting is the sole explanation for AI value realisation, nor that every AI-enabled activity requires the same degree or form of Botsitting. Organisational value continues to depend upon numerous complementary influences, including governance arrangements, organisational capabilities, data quality, leadership, technology design, institutional context, and user competencies14,15. Botsitting should therefore be understood as one complementary element within a broader value realisation process rather than as an independent source of organisational value.

As emerging evidence demonstrates, a significant proportion of AI-enabled productivity gains is consumed by downstream verification and rework activities. We therefore argue that Botsitting effort should be estimated and incorporated as an overhead cost in the return-on-investment (ROI) calculations of AI initiatives, rather than being left unaccounted for, as is typically the case. This is warranted because, wherever AI-generated output is inexpensive and rapid to produce but difficult, ambiguous, expertise-intensive, or time-consuming to verify, the human labour required to render that output trustworthy can absorb part or all of the expected productivity gain59. Accordingly, the expected Botsitting effort should be empirically estimated during the pilot or proof-of-concept stage, which is the phase in which such hidden effort first becomes observable, and then formally incorporated into ROI calculation models so that the reported returns reflect the true cost of producing trustworthy AI output. Although AI may accelerate initial output generation, the human effort required to verify and refine that output can absorb part or all of the expected productivity gain65.

The conceptualisation developed in this section carries three broader implications, developed further in Section 8: it explains why increasing AI capability redistributes rather than eliminates human work; it explains why practitioner communities have recognised Botsitting as AI has become more deeply embedded in organisational work; and it establishes a conceptual bridge connecting concept development, AI-enabled work, IT business value, hidden human labour, and human-AI collaboration around AI value realisation. This provides the foundation for the integrated conceptual model developed in the following section.

A Conceptual Model of Botsitting and AI Value Realisation

Presenting the Conceptual Model

The next stage is to explain how the conceptual elements established in Section 4 relate to one another within the broader process of AI value realisation. Conceptual models contribute to theory development by making such relationships explicit and providing a coherent structure through which emerging concepts can be understood and subsequently examined empirically66,29. Accordingly, Figure 1 presents an integrated conceptual model explaining how Botsitting contributes to AI value realisation.

Figure 1 begins with AI capability, representing the technological capabilities of contemporary AI systems to generate analyses, recommendations, content, decisions, and increasingly autonomous actions8. These capabilities create AI-generated potential, providing organisations with opportunities to improve decision making, productivity, innovation, and organisational performance. However, the existence of technological capability and AI-generated potential does not guarantee that organisational value will be realised. Between AI-generated potential and realised organisational value lies the AI Value Realisation Gap, representing the distinction between technological opportunity and practically realised organisational outcomes.

Within this broader value realisation process, Botsitting is positioned as an emerging form of value-enabling hidden human labour that contributes to narrowing the AI Value Realisation Gap. Rather than functioning independently of AI capability, Botsitting complements it by providing the human judgement, contextual interpretation, verification, refinement, supervision, and organisational integration required to transform AI-generated potential into reliable and organisationally valuable outcomes. The model therefore conceptualises Botsitting as one complementary pathway through which organisations realise value from AI-enabled work.

The model also distinguishes the underlying concept of Botsitting from its observable manifestations. Activities such as supervision, monitoring, verification, interpretation, AI-based refinement, prompt engineering, manual correction, escalation, and quality assurance are included because they illustrate how Botsitting becomes visible in organisational practice, not because they define the concept. Figure 1 should therefore be read as the visual synthesis of the conceptualisation developed in this paper rather than as a separate theoretical contribution.

Figure 1. Botsitting as the Value-Enabling Bridge Between AI-Generated Potential and Realised Value (Note: the Botsitting-realised value relationship holds where the pace, scope, and transparency of AI-generated output remain within practical human review capacity; see Section 5.3 and Proposition 7)

Interpreting the Conceptual Model

The central relationship represented in the model is the distinction between AI-generated potential and realised organisational value. Although increasingly capable AI systems generate significant technological opportunities, organisational value is realised only when those capabilities are complemented by appropriate organisational and human contributions14,67,15. Within this broader process, Botsitting is conceptualised as one form of value-enabling hidden human labour that contributes to narrowing the AI Value Realisation Gap.

Viewed collectively, the conceptual model contributes in three important respects. First, it integrates the conceptualisation developed throughout this paper into a coherent explanatory framework linking AI capability with realised organisational value. Second, it provides a theoretically grounded explanation for why increasing AI capability frequently redistributes rather than eliminates complementary human work. Third, it establishes a conceptual foundation upon which future empirical research can examine, refine, extend, or challenge the relationships proposed in this paper across different organisational contexts.

Boundary Conditions of the Model

The relationships depicted in Figure 1 should not be read as implying that Botsitting can scale indefinitely alongside AI capability. Human-factors research on automation68 has long identified a structural tension between increasing system capability and the practical possibility of effective human oversight. Bainbridge69 observed that the more reliable an automated system becomes, the less practised its human overseers become at detecting the rarer occasions when it fails, precisely when their judgement matters most. Parasuraman and Manzey70 extend this point, showing that complacency and attentional narrowing increase, rather than decrease, as system reliability improves. Agentic AI sharpens this tension: where systems can initiate actions across organisational processes faster than they can be reviewed, Botsitting becomes harder to perform in any meaningful sense even where formal supervisory responsibility remains assigned to named individuals4. Under these conditions, human oversight risks becoming what Elish71 terms a “moral crumple zone”: a locus of formal accountability without a corresponding practical capacity to exercise it. Schoenherr and Michael9 make a related argument, observing that systems expected to remain viable as they evolve require both design principles that anticipate successive stages of adaptation and governance arrangements holding responsiveness and accountability in balance. The conceptual model therefore holds most clearly where the pace, scope, and transparency of AI-generated outputs remain within organisational review capacity; beyond that threshold, a Botsitting deficit is better understood as a system design problem than as a resourcing gap.

At the same time, future advances in explainable AI, AI assurance technologies, automated verification tools, and adaptive governance mechanisms may reduce the amount of Botsitting required for particular tasks. Such developments, however, are more likely to change the nature and distribution of Botsitting than eliminate the need for human judgement, accountability, and contextual interpretation altogether.

The conceptual relationships represented in Figure 1 thus provide the basis for the emerging research propositions developed in the following section. These propositions should not be interpreted as empirical hypotheses. Rather, they articulate theoretically reasoned relationships derived from the conceptual model and provide an initial foundation for the cumulative development of Botsitting research.

Emerging Research Propositions

The conceptual model developed in the preceding section establishes an initial theoretical explanation of how Botsitting contributes to AI value realisation. Building upon this model, the following propositions articulate the principal theoretical relationships emerging from the conceptualisation. They do not extend the conceptual architecture developed in this paper, and they are not empirical hypotheses. Rather, they are theoretically reasoned statements about relationships implied by the model30,26,72,29. The empirical programme through which they might be examined is set out separately in Section 7.

Emergence of Botsitting

The conceptual model suggests that the emergence of Botsitting is closely associated with the increasing integration of AI into organisational work. As organisations embed increasingly capable AI systems within operational and knowledge-intensive processes, they also become more dependent upon complementary human judgement, contextual interpretation, validation, and oversight. Rather than eliminating human work, AI frequently redistributes it by creating new forms of complementary labour that enable AI-generated potential to become organisationally valuable10,7,8.

This complementary human role increasingly depends on meta-skills that enable employees to effectively work with AI systems, including the ability to critically assess AI-generated outputs, provide relevant contextual knowledge, and refine AI responses using domain expertise73,74,75. These capabilities are central to Botsitting because they enable organisations to transform AI-generated outputs from raw AI capabilities into reliable, contextually appropriate, and actionable organisational outcomes.

Proposition 1: As AI becomes increasingly embedded within organisational work, the need for Botsitting is likely to become increasingly salient because AI-generated potential frequently requires complementary human labour to achieve reliable, contextually appropriate, and organisationally valuable outcomes.

Botsitting and AI Value Realisation

Consistent with the IT business value literature, realised value depends upon complementary organisational resources, governance arrangements, work practices, and human capabilities13,67,14,15. Prior research on IT investments similarly emphasises that technological implementation alone does not guarantee the realisation of expected business benefits. Peppard et al.76 argue that value emerges only when technological capabilities are combined with complementary business changes, new work practices, and clear accountability for delivering outcomes. Extending this perspective to AI, the conceptual model identifies Botsitting as one form of value-enabling hidden human labour that contributes to narrowing the AI Value Realisation Gap.

Proposition 2: Botsitting contributes to narrowing the AI Value Realisation Gap by enabling AI-generated potential to be transformed into reliable, contextually appropriate, and organisationally valuable outcomes.

The value of Botsitting depends not only on the presence of human oversight but on how effectively that labour complements AI-generated output. Efficient Botsitting concentrates on the decisive evidence, criteria, and risk indicators; inefficient Botsitting reconstructs the complete analysis because the AI output is insufficiently transparent or traceable59.

Dynamic Manifestations of Botsitting

While technological capability and organisational AI maturity will continue to evolve, the underlying concept of Botsitting is expected to remain comparatively stable because it is defined by its essential nature rather than by any particular activity26,28,29. Current manifestations such as prompt refinement, verification, supervision, interpretation, and quality assurance are therefore contemporary expressions of the concept. Future technological advances may substantially alter how Botsitting is performed, but they are unlikely to eliminate the need for human oversight, contextual judgement, accountability, and organisational coordination.

Proposition 3: The manifestations of Botsitting are expected to evolve alongside advances in AI capability, while its ontological identity as an emerging form of value-enabling hidden human labour remains comparatively stable.

Multi-Level Nature of Botsitting

The conceptual model positions Botsitting as a phenomenon operating across individuals, teams, organisations, and society at large. While Botsitting is frequently observed through individual activities such as reviewing AI-generated outputs or validating recommendations, these activities are embedded within broader organisational processes involving collaboration, governance, shared decision making, and quality assurance. This perspective is consistent with the broader literature demonstrating that organisational value emerges through interactions across multiple levels of analysis77,9 rather than through isolated individual behaviour14,15.

Proposition 4: Botsitting operates across individuals, teams, organisations, sectors, and society, although its manifestations, practices, and organisational implications are likely to differ across these levels of analysis.

Evolution of Botsitting

Botsitting should not be regarded as a temporary response to the limitations of contemporary AI systems. As AI systems become increasingly capable and autonomous, some current manifestations may diminish while new forms of complementary human labour emerge in response to changing organisational and technological conditions10,8. We expect this trend to continue, with managing farms of bots taking on a more literal meaning and tools created to assist with “virtual herding” rather than conventional management.

Proposition 5: Advances in AI capability are likely to transform rather than eliminate Botsitting by changing the forms through which value-enabling hidden human labour is expressed.

Recognition of Botsitting as an Organisational Capability

A central argument developed throughout this paper is that Botsitting frequently remains under-recognised within prevailing assumptions concerning AI productivity and automation. Yet the conceptual model suggests that this complementary human labour contributes directly to narrowing the AI Value Realisation Gap by supporting the transformation of AI-generated potential into realised organisational value. Accordingly, enterprises should recognise Botsitting as actual work and allocate time for it78. Organisations that fail to recognise this contribution may underestimate an important component of successful AI implementation. Conversely, organisations that explicitly acknowledge and support Botsitting are likely to be better positioned to realise sustained value from AI-enabled work.

Proposition 6: Organisations that explicitly recognise, support, and manage Botsitting are likely to be better positioned to realise value from AI-enabled work than organisations that continue to treat Botsitting as invisible labour.

This recognition is necessary but not sufficient: as Section 5.3 argues, beyond a certain threshold of AI speed and autonomy, formally recognising Botsitting without redesigning workflows to keep pace risks formalising oversight in name only.

Boundary Conditions of Botsitting

Section 5.3 identified the conditions under which the conceptual model ceases to hold. Where the pace, scope, and transparency of AI-generated output exceed organisational review capacity, assigning supervisory responsibility does not by itself create the capacity to exercise it, and a Botsitting deficit becomes a problem of system design rather than of resourcing.

Proposition 7: As AI systems become faster, more autonomous, and more deeply networked across organisational processes, the practical capacity to perform effective Botsitting is likely to diminish even as its formal organisational importance increases, widening the gap between nominal human oversight and its meaningful exercise.

Collectively, these seven propositions translate the conceptual model into an initial programme of theoretical enquiry. Rather than representing definitive theoretical claims, they articulate the principal relationships emerging from the conceptualisation developed in this paper. The following section sets out the empirical research agenda through which these relationships might be examined, refined, or challenged.

Future Research Agenda

The conceptualisation, conceptual model, and propositions developed in the preceding sections establish an initial theoretical foundation for Botsitting. Concept development, however, represents the beginning rather than the culmination of scholarly enquiry. Concepts mature through continuing theoretical refinement, empirical investigation, and methodological development as evidence accumulates across different organisational contexts26,58,29. This section sets out the empirical agenda corresponding to the propositions advanced in Section 6, organised into seven themes.

Given the multi-level nature of Botsitting, future research will benefit from methodological pluralism. Qualitative case studies may illuminate how Botsitting emerges in organisational settings, while surveys, experiments, longitudinal studies, and design science research can collectively examine its antecedents, mechanisms, outcomes, and evolution across different organisational contexts.

Refining the Conceptualisation

The conceptualisation advanced in this paper should be regarded as an initial theoretical foundation rather than a definitive representation of Botsitting. Future research should examine the robustness of the proposed conceptual boundaries, ontological identity, and defining attributes across different organisational contexts, industries, and national settings. Such work will strengthen conceptual clarity while supporting cumulative theory development79. Illustrative research questions include:

·       How should the conceptual boundaries of Botsitting evolve as AI technologies become increasingly autonomous?

·       Which defining attributes remain stable across different organisational contexts?

·       How does Botsitting differ from related forms of human work associated with AI-enabled systems?

·       How is Botsitting recognised in the design of new business processes and digital transformation initiatives in organisations?

·       How prominent is Botsitting in non-digital, AI-assisted human workflows?

Understanding Manifestations

A central argument advanced in this paper is that Botsitting should be distinguished from the manifestations through which it becomes observable. Future empirical research should therefore investigate how Botsitting is expressed across occupations, professions, industries, and organisational settings characterised by different levels of AI maturity, while preserving the distinction between manifestations and the underlying concept. Two gaps are immediately apparent. First, the existing evidence base concerns digital workers, that is, those who work primarily at a laptop or desktop21, leaving Botsitting among frontline, manual, and hands-on workers largely unexamined. Second, it remains unclear whether Botsitting generates sufficient complementary work to offset the labour-market effects of agentic AI adoption, which prior research suggests may contribute to significant workforce displacement in particular industries and occupations80. Illustrative research questions include:

·       How do the manifestations of Botsitting differ across organisational, sectoral and societal contexts?

·       Which manifestations become more or less prominent as AI capability evolves?

·       How do generative AI and agentic AI reshape the manifestations of Botsitting?

Under what circumstances should a human be required to remain in the loop rather than responsibility being devolved to an agent?

Measuring Botsitting

Botsitting can have consequences that are not captured by established productivity measures and therefore requires new measurement approaches. The continued development of Botsitting as a scholarly concept ultimately requires reliable approaches to empirical measurement79. This should include measures of Botsitting outcomes and their impact on value. Future research should develop qualitative, quantitative, and mixed-method approaches capable of operationalising Botsitting across different levels of analysis while maintaining consistency with the conceptualisation advanced in this paper. Illustrative research questions include:

·       How can Botsitting be operationalised and measured?

·       Which dimensions best capture the construct?

·       How should Botsitting be measured across individuals, teams, organisations, sectors, and society?

Botsitting and AI Value Realisation

The conceptual model positions Botsitting as one complementary mechanism contributing to AI value realisation. Future research should examine the organisational conditions under which it contributes most effectively to narrowing the AI Value Realisation Gap, together with the complementary capabilities that strengthen or constrain this relationship67. Prior research on data-driven and algorithmic technologies shows that value creation is not automatic but requires complementary work, resources, and capabilities beyond the technology itself81,82,83, and that organisations require technical expertise, domain knowledge, and digital capabilities to realise value from AI-enabled systems84. Relatedly, Bansal et al.85 show that AI explanations shape complementary team performance by affecting how human effort and judgement are coordinated with AI output rather than by simply increasing trust. A specific open question follows from the Work AI Index finding that digital workers spend almost as much time Botsitting as generating AI output: it remains unestablished whether that time reflects limitations in AI capability or limited worker understanding of how Botsitting is performed effectively. Research in this theme should also address how organisations avoid workslop and botshitting. Illustrative research questions include:

·       Under what conditions does Botsitting contribute most effectively to AI value realisation?

·       How does Botsitting influence decision quality, organisational trust, productivity, and innovation?

·       Which organisational capabilities strengthen or weaken these relationships?

·       Does time spent Botsitting reflect limitations in AI capability or limited worker understanding of effective verification?

Multi-Level Perspectives

Botsitting is conceptualised as a phenomenon operating across individuals, teams, organisations, sectors, and society. Future research should therefore adopt multi-level perspectives77 examining interactions across these levels rather than treating Botsitting exclusively as an individual behaviour. At the individual level, prior research indicates that AI and algorithms complement domain expertise, creating greatest value when algorithmic literacy is broadly diffused among workers86; how algorithmic literacy and domain expertise shape the capacity to perform Botsitting effectively, and how workers should be trained for it, is a central open question. At the team and organisational levels, studies should examine how collaboration mechanisms, scenario-based training, governance structures, and leadership practices shape the coordination and institutionalisation of Botsitting. Drawing on Cooper and Zmud's87 IT implementation model, which explains how technologies progress from initial adoption to routinisation and infusion, research could examine whether Botsitting follows a similar trajectory from informal and invisible practice to institutionalised organisational capability. A further avenue concerns how Botsitting practices travel, or fail to travel, across sectors that differ markedly in regulatory exposure, risk tolerance, and professional norms. Illustrative research questions include:

·       How do individual, team, organisational, sectoral, and societal manifestations of Botsitting interact in shaping the realisation of AI-enabled organisational value?

·       How should training and life-long learning initiatives be rolled out to support AI-enablement in all aspects of work and life?

·       Which governance arrangements and leadership practices support effective Botsitting?

·       How does organisational culture influence the development of Botsitting?

·       What conditions are required for multi-stakeholder Botsitting across and beyond organisational boundaries?

Evolution of Botsitting

Botsitting should not be viewed as a static organisational phenomenon. Rather, it is expected to evolve alongside continuing advances in AI capability and organisational AI maturity. Longitudinal and comparative research will therefore be essential for understanding how the manifestations of Botsitting change while its underlying conceptual identity remains comparatively stable. As agentic AI integrates planning, decision-making, memory, and autonomous action across organisational processes88,89, a central question is whether human involvement shifts from task-level verification and refinement toward workflow-level supervision, coordination, exception management, and accountability. Illustrative research questions include:

·       How will Botsitting evolve as AI systems become increasingly adaptive through semi-autonomous or autonomous operation?

·       What will Botsitting capability maturity models look like?

·       Which manifestations are likely to diminish, and which new manifestations are likely to emerge as agents beget agents and become empowered to make adaptive changes in socio-technical systems?

·       Does Botsitting represent a transitional phenomenon or an enduring characteristic of AI-enabled work?

Governance, Regulation, and Assurance

The boundary conditions identified in Section 5.3 suggest a further research theme distinct from those above: how governance and regulatory frameworks should treat Botsitting once it is recognised as the mechanism through which formal human-oversight obligations are actually discharged. Existing regulatory frameworks, including the EU AI Act, and adjacent national AI strategy and assurance frameworks such as the UK's AI for Science Strategy and UK Research and Innovation's (UKRI) AI Research and Innovation Strategic Framework, presuppose that human oversight of AI-influenced decisions is practically exercisable; whether this assumption holds in practice, and under what conditions it breaks down, remains an open empirical question. Parallel work on public-sector algorithmic governance similarly identifies a non-delegable core of judgement that cannot be transferred to automated systems without eroding accountable authority90, raising the question of how organisations identify which oversight obligations fall within that core. The concern is reinforced by evidence that AI explanations may increase reliance on AI recommendations without improving decision quality, as users reduce independent verification91, and by research on algorithm aversion and algorithm appreciation showing that users may become overly sceptical after observing errors92 or place excessive confidence in algorithmic judgement93. Both tendencies distort the level of verification applied to AI output. Whether interface-level interventions such as cognitive forcing functions can restore it94 is a further open question. Illustrative research questions include:

·       At what point does mandated human oversight of AI systems become a formal requirement without a corresponding practical capacity to fulfil it, and how would organisations or regulators detect it?

·       How do existing and emerging AI governance frameworks (e.g., the EU AI Act, sector-specific regulatory guidance, professional standards bodies, and transparency mechanisms such as the UK’s Algorithmic Transparency Recording Standard95) implicitly assume a level of Botsitting capacity, and how could this be made explicit in audit or compliance requirements?

·       What organisational or regulatory mechanisms could verify that human oversight of AI-generated outputs is genuinely exercised, rather than merely nominally assigned?

·       How should public-sector organisations, which face distinct accountability and transparency obligations such as those underpinning the UK's Algorithmic Transparency Recording Standard, resource and govern Botsitting differently from private-sector firms?

·       How can enterprise Botsitting be ensured, as opposed to reliance on public-facing GPTs and LLMs in private or public institutions, to avoid data leakage and maintain risk-minimisation strategies?

Collectively, these seven research themes establish a coherent programme for the continued development of Botsitting research. More broadly, they demonstrate how the conceptualisation advanced in this paper extends beyond introducing a new scholarly concept by providing a foundation for future research on AI value realisation, human-AI collaboration, and organisational value creation. The following section considers the broader theoretical and practical implications of this conceptualisation.

Theoretical and Practical Implications

Theoretical Implications

The conceptualisation developed in this paper has implications extending beyond introducing Botsitting as an emerging scholarly concept. More fundamentally, it contributes to understanding how AI-generated capability becomes realised organisational value, providing a foundation for future theoretical development across IS, management, and organisational research.

First, the conceptualisation contributes to research on AI-enabled work by shifting attention from AI capability alone towards the complementary human labour required to realise organisational value. Existing research has significantly advanced the understanding of AI capabilities and their organisational applications61,2,3. The present study complements this literature by explaining why increasing AI capability frequently redistributes rather than eliminates human work and by identifying Botsitting as one important form of complementary labour supporting AI-enabled work.

Second, the paper extends the IT business value literature by providing a more explicit explanation of how AI-generated potential becomes realised organisational value. While previous research has consistently demonstrated that technological capability alone does not guarantee organisational value13,67,14,15, the conceptualisation advanced here identifies Botsitting as one complementary mechanism contributing to this transformation. In doing so, it enriches existing explanations of organisational value creation without replacing them. In this respect, Botsitting suggests an intriguing parallel with longer-standing debates concerning the relationship between labour and value96, highlighting how human labour may be reconfigured rather than eliminated in increasingly AI-enabled environments.

Third, the paper contributes to concept development by demonstrating how an emerging organisational phenomenon recognised initially within practitioner and professional discourse can be transformed into a theoretically grounded scholarly concept. The distinction developed throughout the manuscript between the organisational phenomenon, the practitioner terminology, and the scholarly concept provides a structured approach that may also inform the conceptualisation of other emerging AI-related organisational phenomena.

Fourth, the conceptualisation contributes to scholarship on human-AI collaboration by identifying one specific form of complementary human labour through which human expertise and AI capability jointly contribute to organisational value7. Rather than positioning AI and humans as competing alternatives, the conceptual model emphasises that successful AI-enabled work increasingly depends upon their complementary interaction.

Fifth, the conceptualisation contributes to multi-level research on digital transformation by showing that AI value realisation is not only an individual-level activity but an organisational phenomenon involving interactions among employees, teams, and governance structures. Existing research has examined technology adoption and use primarily from either individual97 or organisational98 perspectives. By conceptualising Botsitting as a phenomenon that emerges across multiple levels of analysis, this paper provides a foundation for examining how individual expertise, team coordination, and organisational capabilities collectively shape AI-enabled value creation.

Finally, the conceptualisation connects directly to the emerging governance architecture surrounding institutional AI use. The Royal Society’s99 Science in the Age of AI report, the UK Government’s100 AI for Science Strategy, and UKRI’s101 AI Research and Innovation Strategic Framework each treat AI as core infrastructure requiring explicit standards of verification, documentation, and human accountability, rather than as a capability whose responsible use can be assumed. The EU AI Act similarly imposes obligations that presuppose identifiable human oversight of AI-influenced decisions102. Read against this backdrop, Botsitting is not simply an organisational curiosity; it is frequently the unnamed labour through which such assurance obligations are actually discharged in practice.

Practical Implications

The conceptualisation also has important implications for organisations seeking to realise value from AI-enabled work. A central argument developed throughout this paper is that successful AI implementation is not solely a technological challenge. Rather, it is equally an organisational and human challenge because AI-generated potential frequently depends upon complementary human labour before it becomes realised organisational value. Recognising Botsitting therefore encourages organisations to adopt a more comprehensive understanding of AI implementation that extends beyond technological capability alone.

For organisations and senior decision makers, the conceptualisation suggests that the complementary human labour accompanying AI-enabled work should be recognised as a legitimate organisational capability rather than as incidental or invisible work. Doing so may contribute to more realistic expectations regarding AI implementation, resource allocation, governance, workforce planning, and organisational capability development. It also requires organisations to clearly define employees' roles and responsibilities in Botsitting so that human oversight becomes a value-creating activity rather than merely a compliance function. If employees lack a clear understanding of their role or fear that AI mistakes could be attributed to them, they may become rigid and overly vigilant, focusing on avoiding errors rather than exercising independent judgement103. Research indicates that under increased time pressure, employees tend to over-rely on AI models (leading to workslop), while sufficient observation time helps employees form independent judgements and “de-anchor” from AI recommendations (leading to high-quality final output)78. Organisations should therefore design Botsitting practices that enable employees to apply their expertise to improve AI-enabled decisions and maximise organisational value, rather than simply acting as compliance checkers.

For managers, the conceptualisation highlights the importance of designing work practices that support effective collaboration between AI systems and human expertise16. Rather than assuming that increasing AI capability progressively eliminates human involvement, organisations should recognise that complementary forms of human judgement, contextual interpretation, verification, and quality assurance remain essential for achieving reliable organisational outcomes. Botsitting must be acknowledged, planned for, and appropriately measured. However, Botsitting should not be measured merely based on the number of hours spent by a worker for reviewing an AI output. Two workers may spend equal time checking AI-generated output, but one may face substantially greater verification complexity for the same AI-executed workflow or may provide substantially higher-quality final output for the same-quality AI-generated output. Therefore, Botsitting should be measured across four dimensions:

·       Botsitting volume: how much verification must be performed

·       Botsitting complexity: how difficult each verification activity is

·       Botsitting intensity: how deeply the worker investigates the AI output

·       Botsitting effectiveness: whether the verification actually identifies errors and improves quality

For employees, the conceptualisation suggests that future professional capability will increasingly depend upon skills extending beyond operating AI systems themselves. Critical evaluation, contextual judgement, interpretation, verification, refinement, and responsible AI use are likely to become increasingly important components of professional expertise as AI becomes more deeply embedded within organisational work.

The conceptualisation also has implications for AI developers, technology providers, and policy makers. Future AI systems should be designed not only to increase technological capability but also to support effective human-AI collaboration61. Likewise, organisational policies and responsible AI frameworks should explicitly recognise the complementary human labour through which AI-generated potential becomes organisationally valuable. This has particular force for policy makers: regulatory and governance frameworks that mandate human oversight of AI systems, including emerging regimes such as the EU AI Act, should explicitly account for the organisational capacity, training, and time required to perform that oversight, rather than assuming it will occur automatically. Doing so will contribute to a more balanced understanding of AI-enabled work and support more effective AI governance, workforce development, and organisational value creation.

Finally, as both AI systems and the humans using them advance, Botsitting effort should progressively decrease. Better AI models will improve output quality through external technological advancement, thereby reducing the extent of Botsitting effort needed. Improved system design will further reduce Botsitting through organisation-level interventions. Organisational learning and worker training will increase the effectiveness of Botsitting at the team and individual levels, which will also reduce Botsitting effort. However, Botsitting effort will not disappear entirely. Consistent with the principle of diminishing marginal returns, successive technological improvements may continue to generate benefits, but the incremental gains derived from each additional improvement are likely to decline104. So, a minimum level of Botsitting effort will persist even in an idealised human-AI work environment. Further, a fourth dimension will exert a countervailing effect by increasing Botsitting effort. Workers will continue to apply their individual judgement, experience, values, style, and perspective when refining AI-generated outputs. Prior research indicates that human expertise, agency, personalised direction, and control remain important determinants for enhancing quality of outcomes produced through human–AI co-creation105,106,107. Consequently, this may lead workers to produce differentiated outcomes even when their interactions originated from similar or potentially identical AI-generated outputs. This overall conceptualisation is visually presented in Figure 2.

Figure 2. Factors Reducing Botsitting Effort and the Countervailing Effect of Human Differentiation.

Botsitting is also likely to vary according to the scope and configuration of AI adoption within and beyond the organisation, which can be grouped into three broad tiers of practice. At the level of individual tool use, adoption may take the form of AI prompting, where individuals interact with generative AI tools on an ad hoc basis to produce discrete outputs such as text, code, or analysis, with Botsitting confined to the review and refinement of individual prompt–response cycles, or AI vibe coding, in which AI systems generate or co-generate functional code or artefacts based on iterative, conversational human direction, requiring Botsitting activities such as validating logic, testing outputs, and correcting errors introduced by the AI. At the level of organisational processes, adoption may involve human–AI business process semi-automation, where AI is embedded within existing workflows to automate discrete tasks or decisions while humans retain responsibility for oversight, exception handling, and quality assurance at defined checkpoints, or full adaptive AI with humans in the loop, in which AI systems operate with greater autonomy across end-to-end processes, dynamically adapting to changing conditions, while humans remain positioned to intervene, override, or redirect AI behaviour in real time. Finally, at the level of cross-sector coordination, organisations may participate in adaptation based on common governance and standards frameworks, whereby shared regulatory requirements, industry standards, and interoperability protocols enable Botsitting practices to be coordinated, benchmarked, and mutually recognised across organisational and sectoral boundaries. These three tiers do not represent a fixed sequence that organisations necessarily progress through; rather, an organisation may adopt any combination of these modes depending on context, need, and capability. All three tiers, in turn, operate within a broader societal frame of public trust and legitimacy: the way AI-enabled work is monitored, corrected, and accounted for at the individual, organisational, and sectoral levels collectively shapes, and is shaped by, societal expectations about the reliability and trustworthiness of AI-enabled work (Figure 3).

Figure 3. Modes of AI Adoption Relevant to Botsitting, Grouped by Level of Analysis.

Note. Individual tool use, organisational processes, and cross-sector coordination represent distinct but non-sequential tiers of AI adoption. Organisations may adopt any combination of these modes depending on context, need, and capability. All three tiers operate within, and in turn shape, a broader societal frame of public trust and legitimacy in AI-enabled work.

Conclusion

The central insight emerging from this paper is that AI value realisation within the organisation depends substantially on the human effort of Botsitting, alongside the complementary organisational resources and capabilities identified in the IT business value literature. The conceptualisation developed in this study suggests that this transformation depends not only on technological capability but also on augmentative and complementary forms of human labour that frequently remain under-recognised. By conceptualising Botsitting as an emerging form of value-enabling hidden human labour, the paper provides a new theoretical lens for understanding one previously under-conceptualised pathway through which AI-generated potential becomes reliable, contextually appropriate, and organisationally valuable. In doing so, it contributes not only to the emerging scholarship on Botsitting, but more broadly to understanding AI value realisation in contemporary organisations.

As AI systems continue to become more capable, autonomous, and deeply embedded within organisational work, an intriguing question remains. If the next generation of AI can increasingly generate knowledge, decisions, and actions, the AI Value Realisation Paradox raises the question of whether the defining challenge for organisations will be the continued development of AI capability or the understanding, design, and management of the complementary forms of human labour through which that capability becomes meaningful organisational value? The answer to that question may ultimately shape not only the future of Botsitting research, but also how scholarship understands value creation in the age of artificial intelligence.

References

1. Brynjolfsson, E., & McAfee, A. (2014). The second machine age: Work, progress, and prosperity in a time of brilliant technologies. W. W. Norton & Company.

2. Dwivedi, Y. K., Kshetri, N., Hughes, L., Slade, E. L., Jeyaraj, A., Kar, A. K., Baabdullah, A. M., Koohang, A., Raghavan, V., Ahuja, M., Albanna, H., Albashrawi, M., Al-Busaidi, K. A., Balakrishnan, J., Barlette, Y., Basu, S., Bose, I., Brooks, L., Buhalis, D., … Wright, R. (2023). So what if ChatGPT wrote it? Multidisciplinary perspectives on opportunities, challenges and implications of generative conversational AI for research, practice and policy. International Journal of Information Management, 71, 102642. https://doi.org/10.1016/j.ijinfomgt.2023.102642

3. Feuerriegel, S., Hartmann, J., Janiesch, C., & Zschech, P. (2024). Generative AI. Business & Information Systems Engineering, 66(1), 111–126. https://doi.org/10.1007/s12599-023-00834-7

4. Hughes, L., Dwivedi, Y. K., Malik, T., Shawosh, M., Albashrawi, M. A., Jeon, I., Dutot, V., Appanderanda, M., Crick, T., De’, R., Fenwick, M., Gunaratnege, S. M., Jurcys, P., Kar, A. K., Kshetri, N., Li, K., Mutasa, S., Samothrakis, S., Wade, M., & Walton, P. (2025). AI agents and agentic systems: A multi-expert analysis. Journal of Computer Information Systems, 65(4), 489–517. https://doi.org/10.1080/08874417.2025.2483832

5. Microsoft (2025). 2025 Work Trend Index Annual Report: The year the frontier firm is born. Microsoft. Available at: https://www.microsoft.com/worklab/work-trend-index

6. Maslej, N., Fattorini, L., Perrault, R., Gil, Y., Parli, V., Kariuki, N., Capstick, E., Reuel, A., Brynjolfsson, E., Etchemendy, J., Ligett, K., Lyons, T., Manyika, J., Niebles, J. C., Shoham, Y., Wald, R., Walsh, T., Hamrah, A., Santarlasci, L., Lotufo, J. B., Rome, A., Shi, A., & Oak, S. (2025). Artificial Intelligence Index Report. Stanford Institute for Human-Centered Artificial Intelligence. Available at: https://hai.stanford.edu/ai-index/2025-ai-index-report

7. Dellermann, D., Ebel, P., Söllner, M., & Leimeister, J. M. (2019). The future of human–AI collaboration: A taxonomy of design knowledge for hybrid intelligence systems. Proceedings of the 52nd Hawaii International Conference on System Sciences, 274–283. Available at: https://www.alexandria.unisg.ch/handle/20.500.14171/99162

8. Raisch, S., & Krakowski, S. (2021). Artificial intelligence and management: The automation–augmentation paradox. Academy of Management Review, 46(1), 192–210. https://doi.org/10.5465/amr.2018.0072

9. Schoenherr, J. R., & Michael, K. (2026). Multi-Level Sociotechnical Systems Approach to Human-Centered AI. In: Xu, W. (eds) Handbook of Human-Centered Artificial Intelligence. Springer. https://doi.org/10.1007/978-981-97-8440-0_100-1

10. Jarrahi, M. H. (2018). Artificial intelligence and the future of work: Human–AI symbiosis in organizational decision making. Business Horizons, 61(4), 577–586. https://doi.org/10.1016/j.bushor.2018.03.007

11. MIT NANDA (2025). The GenAI Divide. State of AI in Business 2025. Available at: https://mlq.ai/media/quarterly_decks/v0.1_State_of_AI_in_Business_2025_Report.pdf

12. Dooley, P. C. (2005). The labour theory of value. Routledge.

13. Brynjolfsson, E. (1993). The productivity paradox of information technology. Communications of the ACM, 36(12), 66–77. https://doi.org/10.1145/163298.163309

14. Melville, N., Kraemer, K., & Gurbaxani, V. (2004). Information technology and organizational performance: An integrative model of IT business value. MIS Quarterly, 28(2), 283–322. https://doi.org/10.2307/25148636

15. Kohli, R., & Grover, V. (2008). Business value of IT: An essay on expanding research directions to keep up with the times. Journal of the Association for Information Systems, 9(1), 23–39. https://doi.org/10.17705/1jais.00147

16. Wilson, H. J., & Daugherty, P. R. (2018). Collaborative intelligence: Humans and AI are joining forces. Harvard Business Review, 96(4), 114–123. Available at: https://hbr.org/2018/07/collaborative-intelligence-humans-and-ai-are-joining-forces

17. Duan, Y., Edwards, J. S., & Dwivedi, Y. K. (2019). Artificial intelligence for decision making in the era of Big Data: Evolution, challenges and research agenda. International Journal of Information Management, 48, 63–71. https://doi.org/10.1016/j.ijinfomgt.2019.01.021

18. Dwivedi, Y. K., Hughes, L., Ismagilova, E., Aarts, G., Coombs, C., Crick, T., Duan, Y., Dwivedi, R., Edwards, J., Eirug, A., Galanos, V., Ilavarasan, P. V., Janssen, M., Jones, P., Kar, A. K., Kizgin, H., Kronemann, B., Lal, B., Lucini, B., … Williams, M. D. (2021). Artificial Intelligence (AI): Multidisciplinary perspectives on emerging challenges, opportunities, and agenda for research, practice and policy. International Journal of Information Management, 57, 101994. https://doi.org/10.1016/j.ijinfomgt.2019.08.002

19. Petkovic, D. (2026). GenAI: Our tool or our future master? IEEE Transactions on Technology and Society, 7(1), 15–23. https://doi.org/10.1109/TTS.2025.3603643

20. Niederhoffer, K., Kellerman, G. R., Lee, A., Liebscher, A., Rapuano, K., & Hancock, J. T. (2025). AI-Generated “Workslop” Is Destroying Productivity. Available at: https://hbr.org/2025/09/ai-generated-workslop-is-destroying-productivity

21. Hinds, R., Hoffman, M., Baladi, S., Cao, H., Lee, Y. S., Leonardi, P., Ranganathan, A., Rhymer, J., Rogelberg, S., Sutton, B., & Zhu, Y. (2026). The Work AI Index 2026: Global Botsitting, botshitting, and the hidden human labor of AI at work. Work AI Institute. Available at: https://www.glean.com/work-ai-institute/reports/work-ai-index

22. CIO (2026). The hidden cost of enterprise AI: 6.4 hours a week babysitting bots. Available at: https://www.cio.com/article/4183804/the-hidden-cost-of-enterprise-ai-6-4-hours-a-week-babysitting-bots.html

23. Business Insider (2026). The rise of the "botsitters". Available at:   https://www.businessinsider.com/botsitting-ai-hidden-human-labor-at-work-2026-6

24. TechRadar (2026). Workers are spending hours every week “botsitting” to make sure AI does its job properly. Available at: https://www.techradar.com/pro/workers-are-spending-hours-every-week-botsitting-to-make-sure-ai-does-its-job-properly

25. The Register (2026). Brit workers waste nearly six hours a week “botsitting” AI. Available at: https://www.theregister.com/ai-and-ml/2026/06/10/brit-workers-waste-nearly-six-hours-a-week-botsitting/5253483

26. Wacker, J. G. (2004). A theory of formal conceptual definitions: Developing theory-building measurement instruments. Journal of Operations Management, 22(6), 629–650. https://doi.org/10.1016/j.jom.2004.08.002

27. Suddaby, R. (2010). Editor's comments: Construct clarity in theories of management and organization. Academy of Management Review, 35(3), 346–357. https://doi.org/10.5465/amr.35.3.zok346

28. MacInnis, D. J. (2011). A framework for conceptual contributions in marketing. Journal of Marketing, 75(4), 136–154. https://doi.org/10.1509/jmkg.75.4.136

29. Jaakkola, E. (2020). Designing conceptual articles: Four approaches. AMS Review, 10(1–2), 18–26. https://doi.org/10.1007/s13162-020-00161-0

30. Whetten, D. A. (1989). What constitutes a theoretical contribution? Academy of Management Review, 14(4), 490–495. https://doi.org/10.5465/amr.1989.4308371

31. Dwivedi, Y. K., Jeyaraj, A., Hughes, L., Davies, G. H., Ahuja, M., Albashrawi, M. A., Al-Busaidi, A. S., Al-Sharhan, S., Al-Sulaiti, K. I., Altinay, L., Amalaya, S., Archak, S., Ballestar, M. T., Bhagwat, S. A., Bharadwaj, A., Bhushan, A., Bose, I., Budhwar, P., Bunker, D., … Walton, P. (2024). “Real impact”: Challenges and opportunities in bridging the gap between research and practice – Making a difference in industry, policy, and society. International Journal of Information Management, 78, 102750. https://doi.org/10.1016/j.ijinfomgt.2023.102750

32. Webster, J., & Watson, R. T. (2002). Analyzing the past to prepare for the future: Writing a literature review. MIS Quarterly, 26(2), 13–23. https://doi.org/10.2307/4132319

33. Tangi, L., Rodriguez Müller, A. P., & Janssen, M. (2025). AI-augmented government transformation: Organisational transformation and the sociotechnical implications of artificial intelligence in public administrations. Government Information Quarterly, 42(3), 102055. https://doi.org/10.1016/j.giq.2025.102055

34. Dehouche, N. (2026). Creation, validation, obsolescence: Observed evidence of AI-driven labor market displacement, 2020–2025. Frontiers in Human Dynamics, 8, 1815037. https://doi.org/10.3389/fhumd.2026.1815037

35. Bonfiglioli, A., Crinò, R., Gancia, G., & Papadakis, I. (2025). Artificial intelligence and jobs: Evidence from US commuting zones. Economic Policy, 40(121), 145–194. https://doi.org/10.1093/epolic/eiae059

36. Noy, S., & Zhang, W. (2023). Experimental evidence on the productivity effects of generative artificial intelligence. Science, 381(6654), 187–192. https://doi.org/10.1126/science.adh2586

37. Dell'Acqua, F., McFowland, E., III, Mollick, E. R., Lifshitz-Assaf, H., Kellogg, K., Rajendran, S., Krayer, L., Candelon, F., & Lakhani, K. R. (2026). Navigating the jagged technological frontier: Field experimental evidence of the effects of artificial intelligence on knowledge worker productivity and quality. Organization Science. https://doi.org/10.1287/orsc.2025.21838

38. Vaccaro, M., Almaatouq, A., & Malone, T. (2024). When combinations of humans and AI are useful: A systematic review and meta-analysis. Nature Human Behaviour, 8, 2293–2303. https://doi.org/10.1038/s41562-024-02024-1

39. Manske, F., & Wolf, H. (1989). Design work in change: Social conditions and results of CAD use in mechanical engineering. IEEE Transactions on Engineering Management, 36(4), 282–292. https://doi.org/10.1109/17.40981

40. Ritson, M. (2026). 65% of marketing jobs may not survive AI. Adweek. Available at: https://www.adweek.com/brand-marketing/65-of-marketing-jobs-may-not-survive-ai/

41. Torres, R. (2026). AI triggers hiring shift for Fortune 500. CIO Dive. Available at: https://www.ciodive.com/news/hiring-AI-skills-fortune-500-draup/813011/

42. Tecotzky, A. (2026). Jamie Dimon says JPMorgan will probably hire fewer bankers in the future—and more “AI people”. Business Insider. Available at: https://www.businessinsider.com/jamie-dimon-jpmorgan-ai-bankers-job-loss2026-5

43. Chouchane, A. (2025). How control rooms are changing and the role of AI assistants in CCTV operations. Available at: https://intelexvision.com/how-control-rooms-are-changing-and-the-role-ai-assistants-in-cctv-operations/

44. AxxonSoft (2024). How AI-Powered Security is Transforming CCTV and Surveillance — Challenges and Solutions for Industry Professionals. Available at: https://www.axxonsoft.com/blog/how-artificial-intelligence-is-transforming-cctv-security

45. SafetyScope (2026). AI video analytics vs. traditional CCTV monitoring. Available at: https://safetyscope.eu/compare/ai-video-analytics-vs-traditional-cctv

46. Head, D. W. (2025). Will AI CCTV monitoring replace human security operators? SureSafe. Available at:  https://us.personalalarms.org/will-ai-cctv-monitoring-replace-human-security-operators/

47. Suleman, S. (2026). 10 Ways CCTV Redaction Software Improves Public Safety and Commercial Security. VidizmoAI. Available at: https://vidizmo.ai/blog/cctv-redaction-software

48. Kovacs, E. (2024). Microsoft unveils Copilot Vision AI tool, but highlights security after Recall debacle. Security Week. Available at: https://www.securityweek.com/microsoft-unveils-copilot-vision-ai-tool-but-highlights-security-after-recall-debacle/

49. Michael, K., Abbas, R., & Roussos, G. (2023). AI in cybersecurity: The paradox. IEEE Transactions on Technology and Society, 4(2), 104–109. https://doi.org/10.1109/TTS.2023.3280109

50. BBC (2005). CCTV systems "fail to cut crime." BBC News. Available at: http://news.bbc.co.uk/2/hi/uk_news/england/leicestershire/4294693.stm

51. Daniels, A. K. (1987). Invisible work. Social Problems, 34(5), 403–415. https://doi.org/10.2307/800538

52. Gray, M. L., & Suri, S. (2019). Ghost work: How to stop Silicon Valley from building a new global underclass. Houghton Mifflin Harcourt.

53. Muldoon, J., Graham, M., & Cant, C. (2024). Feeding the machine: The hidden human labour powering AI. New York, USA: Bloomsbury Publishing.

54. Casilli, A. A. (2024). Digital labor and the inconspicuous production of artificial intelligence. arXiv:2410.05910. https://doi.org/10.48550/arXiv.2410.05910

55. Michael, K., Schoenherr, J. R., & Vogel, K. M. (2024). Failures in the loop: Human leadership in AI-based decision-making. IEEE Transactions on Technology and Society, 5(1), 2–13. https://doi.org/10.1109/TTS.2024.3378587

56. Mackworth, N. H. (1948). The breakdown of vigilance during prolonged visual search. Quarterly Journal of Experimental Psychology, 1(1), 6–21. https://doi.org/10.1080/17470214808416738

57. SoftwareSeni (2026). Botsitting and botshitting: The empirical picture from the Work AI Index 2026. Available at: https://www.softwareseni.com/botsitting-and-botshitting-the-empirical-picture-from-the-work-ai-index-2026/

58. Podsakoff, P. M., MacKenzie, S. B., & Podsakoff, N. P. (2016). Recommendations for creating better conceptual definitions in the organizational, behavioral, and social sciences. Organizational Research Methods, 19(2), 159–203. https://doi.org/10.1177/1094428115624965

59. Zhu, L., Lu, Q., Ding, M., Lee, S. U., & Wang, C. (2026). Designing meaningful human oversight in AI. AI and Ethics, 6(3), 286. https://doi.org/10.1007/s43681-026-01147-7

60. Gross, G. (2026). ‘Botsitting’: The AI time-savings killer only governance can stop. CIO. Available at: https://www.cio.com/article/4188575/botsitting-the-ai-time-savings-killer-only-governance-can-stop.html

61. Faraj, S., Pachidi, S., & Sayegh, K. (2018). Working and organizing in the age of artificial intelligence. Information and Organization, 28(1), 62–70. https://doi.org/10.1016/j.infoandorg.2018.02.005

62. Hatton, E. (2017). Mechanisms of invisibility: Rethinking the concept of invisible work. Work, Employment and Society, 31(2), 336–351. https://doi.org/10.1177/0950017016674894

63. Rani, U., Göbel, C., & Dhir, R. K. (2024). Development of artificial intelligence: The role of invisible workers in the AI value chain. United Nations. Available at: https://sdgs.un.org/documents/rani-u-gobel-n-and-dhir-r-development-ai-role-invisible-workers-ai-value-chains-55572

64. Elgendy, I. A., Dwivedi, Y. K., Al-Sharafi, M. A., Hosny, M., Helal, M. Y. I., Crick, T., Hughes, L., Alwahaishi, S., Mahmud, M., Dutot, V., & Al-Busaidi, A. S. (2026). Responsible Vibe Coding: Architecture, Opportunities, and Research Agenda. Journal of Computer Information Systems, 1–19. https://doi.org/10.1080/08874417.2026.2621186

65. Wolfe, R., & Mitra, T. (2024). The impact and opportunities of generative AI in fact-checking. Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency (FAccT '24). https://doi.org/10.1145/3630106.3658987

66. Burton-Jones, A., McLean, E. R., & Monod, E. (2015). Theoretical perspectives in IS research: From variance and process to conceptual latitude and conceptual fit. European Journal of Information Systems, 24(6), 664–679. https://doi.org/10.1057/ejis.2014.31

67. Soh, C., & Markus, M. L. (1995). How IT creates business value: A process theory synthesis. Proceedings of the Sixteenth International Conference on Information Systems, 29–41. http://aisel.aisnet.org/icis1995/4

68. Parasuraman, R., & Riley, V. (1997). Humans and automation: Use, misuse, disuse, abuse. Human Factors, 39(2), 230–253. https://doi.org/10.1518/001872097778543886

69. Bainbridge, L. (1983). Ironies of automation. Automatica, 19(6), 775–779. https://doi.org/10.1016/0005-1098(83)90046-8

70. Parasuraman, R., & Manzey, D. H. (2010). Complacency and bias in human use of automation: An attentional integration. Human Factors, 52(3), 381–410. https://doi.org/10.1177/0018720810376055

71. Elish, M. C. (2019). Moral crumple zones: Cautionary tales in human-robot interaction. Engaging Science, Technology, and Society, 5, 40–60. https://doi.org/10.17351/ests2019.260

72. Gregor, S. (2006). The nature of theory in information systems. MIS Quarterly, 30(3), 611–642. https://doi.org/10.2307/25148742

73. Call, M. L., Jiang, K., & Idso, C. (2026). Star advantage: Employee value creation and capture in the age of artificial intelligence. Human Resource Management, 65(1), 151–167. https://doi.org/10.1002/hrm.70023

74. Krakowski, S., Luger, J., & Raisch, S. (2023). Artificial intelligence and the changing sources of competitive advantage. Strategic Management Journal, 44(6), 1425–1452. https://doi.org/10.1002/smj.3387

75. Choudhury, P., Starr, E., & Agarwal, R. (2020). Machine learning and human capital complementarities: Experimental evidence on bias mitigation. Strategic Management Journal, 41(8), 1381–1411. https://doi.org/10.1002/smj.3152

76. Peppard, J., Ward, J., & Daniel, E. (2007). Managing the realization of business benefits from IT investments. MIS Quarterly Executive, 6(1), 1–11. https://aisel.aisnet.org/misqe/vol6/iss1/3

77. Kozlowski, S. W. J., & Klein, K. J. (2000). A multilevel approach to theory and research in organizations: Contextual, temporal, and emergent processes. In K. J. Klein & S. W. J. Kozlowski (Eds.), Multilevel theory, research, and methods in organizations (pp. 3–90). Jossey-Bass.

78. Romeo, G., & Conti, D. (2026). Exploring automation bias in human–AI collaboration: A review and implications for explainable AI. AI & Society, 41(1), 259–278. https://doi.org/10.1007/s00146-025-02422-7

79. MacKenzie, S. B., Podsakoff, P. M., & Podsakoff, N. P. (2011). Construct measurement and validation procedures in MIS and behavioral research: Integrating new and existing techniques. MIS Quarterly, 35(2), 293–334. https://doi.org/10.2307/23044045

80. Pati, A. K. (2025). Agentic AI: A comprehensive survey of technologies, applications, and societal implications. IEEE Access, 13, 151824–151837. https://doi.org/10.1109/ACCESS.2025.3586076

81. Grønsund, T., & Aanestad, M. (2020). Augmenting the algorithm: Emerging human-in-the-loop work configurations. The Journal of Strategic Information Systems, 29(2), 101614. https://doi.org/10.1016/j.jsis.2020.101614

82. Scheepers, R., Lacity, M. C., & Willcocks, L. P. (2018). Cognitive automation as part of Deakin University’s digital strategy. MIS Quarterly Executive, 17(2), 89–107. https://aisel.aisnet.org/misqe/vol17/iss2/4

83. Watson, H. J. (2017). Preparing for the cognitive generation of decision support. MIS Quarterly Executive, 16(3), 153–169. https://aisel.aisnet.org/misqe/vol16/iss3/3

84. Tarafdar, M., Beath, C. M., & Ross, J. W. (2019). Using AI to enhance business operations. MIT Sloan Management Review, 60(4), 37–44. https://sloanreview.mit.edu/article/using-ai-to-enhance-business-operations/

85. Bansal, G., Wu, T., Zhou, J., Fok, R., Nushi, B., Kamar, E., Ribeiro, M. T., & Weld, D. (2021). Does the whole exceed its parts? The effect of AI explanations on complementary team performance. Proceedings of the 2021 CHI Conference on Human Factors in Computing Systems, 1–16. https://doi.org/10.1145/3411764.3445717

86. Tambe, P. B. (2026). Reskilling the workforce for AI: Domain expertise and algorithmic literacy. Management Science, 72(1).  https://doi.org/10.1287/mnsc.2022.03968

87. Cooper, R. B., & Zmud, R. W. (1990). Information technology implementation research: A technology diffusion approach. Management Science, 36(2), 123–139. https://doi.org/10.1287/mnsc.36.2.123

88. Murugesan, S. (2025). The rise of agentic AI: Implications, concerns, and the path forward. IEEE Intelligent Systems, 40(2), 8–14. https://doi.org/10.1109/MIS.2025.3544940

89. Kshetri, N. (2025). Economics of agentic AI in the health-care industry. IT Professional, 27(1), 14–19. https://doi.org/10.1109/MITP.2025.3529857

90. Engin, Z., Gordon, T., Bastidas, V., Crick, T., Crowcroft, J., Denis, J.-M., Hand, D. J., Maffeo, L., Mökander, J., Ng, I., Nikiforova, A., Quaggiotto, G., Socol de la Osa, D. U., Syler, R., Treleaven, P., & Verhulst, S. (2026). Governing well in the algorithmic age: The foundations of digital statecraft. arXiv:2607.18483. https://doi.org/10.48550/arXiv.2607.18483

91. Lane, J. N., Lin, C., Boussioux, L., Spens, R., Ayoubi, C., Wagh, P., Chen, Y. H., & Wang, P.-H. (2025). The narrative AI advantage? A field experiment on AI-augmented evaluations of early-stage innovations (Harvard Business School Working Paper No. 25-001). Harvard Business School. Available at: https://www.hbs.edu/ris/Publication%20Files/25-001_8ebbe0cb-2a19-453c-9014-1e301e8dd2fb.pdf

92. Dietvorst, B. J., Simmons, J. P., & Massey, C. (2015). Algorithm aversion: People erroneously avoid algorithms after seeing them err. Journal of Experimental Psychology: General, 144(1), 114–126. https://doi.org/10.1037/xge0000033

93. Logg, J. M., Minson, J. A., & Moore, D. A. (2019). Algorithm appreciation: People prefer algorithmic to human judgment. Organizational Behavior and Human Decision Processes, 151, 90–103. https://doi.org/10.1016/j.obhdp.2018.12.005

94. Bučinca, Z., Malaya, M. B., & Gajos, K. Z. (2021). To trust or to think: Cognitive forcing functions can reduce overreliance on AI in AI-assisted decision-making. Proceedings of the ACM on Human-Computer Interaction, 5(CSCW1), 188. https://doi.org/10.1145/3449287

95. UK Government (2023). Algorithmic Transparency Recording Standard. Government Digital Service. Available at: https://www.gov.uk/government/collections/algorithmic-transparency-recording-standard-hub

96. Zhang, F. (2026). The labor theory of value in the era of artificial intelligence and digital platforms: Challenges, innovations, and new mechanisms. Humanities and Social Sciences Communications, 13(1), 717. https://doi.org/10.1057/s41599-026-07030-4

97. Choi, S., Jang, Y., & Kim, H. (2022). Influence of pedagogical beliefs and perceived trust on teachers' acceptance of educational artificial intelligence tools. International Journal of Human–Computer Interaction, 39(4), 1–13. https://doi.org/10.1080/10447318.2022.2049145

98. Simões, A. C., Lucas Soares, A., & Barros, A. C. (2020). Factors influencing the intention of managers to adopt collaborative robots (cobots) in manufacturing organizations. Journal of Engineering and Technology Management, 57, 101574. https://doi.org/10.1016/j.jengtecman.2020.101574

99. Royal Society (2024). Science in the age of AI: How artificial intelligence is changing the nature and method of scientific research. The Royal Society. Available at: https://royalsociety.org/news-resources/projects/science-in-the-age-of-ai/

100. UK Government (2025). AI for Science Strategy. CP 1241. Department for Science, Innovation and Technology. Available at: https://www.gov.uk/government/publications/ai-for-science-strategy/ai-for-science-strategy

101. UK Research and Innovation (2026). UKRI AI Research and Innovation Strategic Framework. Available at: https://www.ukri.org/publications/ukri-artificial-intelligence-research-and-innovation-strategic-framework/

102. European Union (2024). Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Official Journal of the European Union, L 2024/1689. https://eur-lex.europa.eu/eli/reg/2024/1689/oj

103. Colletta, J. (2026). ‘Botsitting’ is wasting more than 6 employee hours a week. HR Executive. Available at: https://hrexecutive.com/botsitting-is-wasting-more-than-6-employee-hours-a-week/

104. Bloom, N., Jones, C. I., Van Reenen, J., & Webb, M. (2020). Are ideas getting harder to find? American Economic Review, 110(4), 1104–1144. https://doi.org/10.1257/aer.20180338

105. Salma, Z., Hijón-Neira, R., & Pizarro, C. (2025). Designing co-creative systems: Five paradoxes in human–AI collaboration. Information, 16(10), 909. https://doi.org/10.3390/info16100909

106. Wang, N., Kim, H., Peng, J., & Wang, J. (2025). Exploring creativity in human–AI co-creation: A comparative study across design experience. Frontiers in Computer Science, 7, 1672735. https://doi.org/10.3389/fcomp.2025.1672735

107. Ren, R., Ma, J., & Luo, J. (2026). Human–AI co-creation: Why, what, and how? Proceedings of the Design Society, 6, 2503–2510. https://doi.org/10.1017/pds.2026.10608

Citation: Dwivedi, Y.K., Crick, T., Michael, K., Samothrakisn, S., Sarker, I., Naeema, R.M., Raman, R., Malik, T., Dutota, V., Jeon, Il, Helal, M.Y.I., 2026, “Botsitting: Conceptualising Value-Enabling Hidden Human Labour in AI-Enabled Work”, Journal of Computer Information Systems, https://doi.org/10.1080/08874417.2026.2749411; https://www.tandfonline.com/doi/full/10.1080/08874417.2026.2749411